Scanning Chromatin: a New Paradigm?*

In recent years, thinking about chromatin function has been dominated by two powerful concepts: the “histone code” (1, 2) and “chromatin remodeling” (3, 4). According to the histone code model, the pattern of covalent histone modifications along a chain of nucleosomes somehow encodes the information that allows site-specific assembly of those DNA-protein complexes that lead to the selective transcription of specific genes. Such complexes may include not only the transcription initiator complex itself but also remodeling factors that are capable of either moving nucleosomes along the DNA or otherwise making the DNAmore accessible to other factors (3, 4). The generality and remarkable diversity of histone modifications, as well as the ubiquity of remodeling factors, argues strongly that these are indeed important aspects of gene regulation, although how specific the “code” may really be has recently been called to question (5). In any event, an implicit assumption that has dominated the field is that chromatin is essentially a passive entity, waiting to be appropriately marked or remodeled to allow it to be transcribed, replicated, or repaired. For example, the various pathways by which a preinitiation complexmight be assembled at a particular promoter have been carefully discussed by Narlikar et al. (6). In none of the pathways considered do the nucleosomes themselves play an active role. If chromatin is indeed passive, a serious question arises as to how specific sites can be recognized for the appropriate marking or remodeling. Many promoters, when inactive, appear to have the binding sites for transcription factors buried in nucleosomes. The conventional wisdom is that chromatin remodeling complexes, perhaps recruited to the site by specific histone modification markers, obligingly shift or “modify” the nucleosomes so as to make such sites accessible. Nevertheless, the primary question of how a hidden location is initially recognized remains unanswered. Very recently, a new view of chromatin behavior has emerged that puts such questions in a very different, and perhaps more illuminating, light. This perspective regards chromatin as a highly dynamic structure, in which binding sites are continually being scanned by nuclear proteins in a random, undirected fashion. In this view, whether or not a gene will be selected for transcription depends on a sequence of dynamic events each with a given probability. There are two necessary prerequisites for such scanning: first, proteins must be free to move about the nucleus, and second, a mechanism must exist to allow occasional exposure of binding sites on all or most nucleosomes. Remarkably, both of these prerequisites turn out to be true.

In recent years, thinking about chromatin function has been dominated by two powerful concepts: the "histone code" (1,2) and "chromatin remodeling" (3,4). According to the histone code model, the pattern of covalent histone modifications along a chain of nucleosomes somehow encodes the information that allows site-specific assembly of those DNA-protein complexes that lead to the selective transcription of specific genes. Such complexes may include not only the transcription initiator complex itself but also remodeling factors that are capable of either moving nucleosomes along the DNA or otherwise making the DNA more accessible to other factors (3,4). The generality and remarkable diversity of histone modifications, as well as the ubiquity of remodeling factors, argues strongly that these are indeed important aspects of gene regulation, although how specific the "code" may really be has recently been called to question (5). In any event, an implicit assumption that has dominated the field is that chromatin is essentially a passive entity, waiting to be appropriately marked or remodeled to allow it to be transcribed, replicated, or repaired. For example, the various pathways by which a preinitiation complex might be assembled at a particular promoter have been carefully discussed by Narlikar et al. (6). In none of the pathways considered do the nucleosomes themselves play an active role.
If chromatin is indeed passive, a serious question arises as to how specific sites can be recognized for the appropriate marking or remodeling. Many promoters, when inactive, appear to have the binding sites for transcription factors buried in nucleosomes. The conventional wisdom is that chromatin remodeling complexes, perhaps recruited to the site by specific histone modification markers, obligingly shift or "modify" the nucleosomes so as to make such sites accessible. Nevertheless, the primary question of how a hidden location is initially recognized remains unanswered.
Very recently, a new view of chromatin behavior has emerged that puts such questions in a very different, and perhaps more illuminating, light. This perspective regards chromatin as a highly dynamic structure, in which binding sites are continually being scanned by nuclear proteins in a random, undirected fashion. In this view, whether or not a gene will be selected for transcription depends on a sequence of dynamic events each with a given probability. There are two necessary prerequisites for such scanning: first, proteins must be free to move about the nucleus, and second, a mechanism must exist to allow occasional exposure of binding sites on all or most nucleosomes. Remarkably, both of these prerequisites turn out to be true.

Protein Mobility in the Nucleus
The first experimental evidence to indicate that the protein composition of chromatin might not be static came from in vitro studies in the laboratory of J. Thomas, in the early 1980s (7,8). Although transfer of histone H1 from chromatin to free nucleic acids had been demonstrated earlier (see, for example, Ref. 9), Thomas's experiments, using radiolabeled histones, clearly showed H1 transfer between chromatin fragments. Furthermore, the process was fast, transfer occurring in less than 1 h ("possibly much less"; Ref. 7). However, the methods available at the time could neither allow a precise measure of the exchange rate nor demonstrate exchange in vivo. Perhaps for these reasons, this seminal work was largely unappreciated for nearly 2 decades.
The emergence of two new techniques, fluorescence recovery after photo-bleaching (FRAP) 3 and single molecule fluorescence imaging, have dramatically demonstrated protein dynamics in the nucleus. Most experiments to date have utilized the FRAP technique, with proteins to be studied genetically fused to the green fluorescent protein (GFP) and expressed in the living cell (10). However, direct fluorescent imaging of individual proteins in the nucleus is now possible and seems ultimately to promise more detailed information concerning the mechanics of protein motility in situ (11). The basic observation in the FRAP experiment is the kinetics of fluorescence regain in a small volume (of the nucleus, for example) that has been photobleached by a highly focused laser pulse. The rate at which fluorescence is regained in the photobleached region is determined by the rate at which fluorescent molecules can move from the surrounding unbleached region back into the bleached region. This, in turn, is determined by the free diffusion rate, modified by the delays each molecule experiences if it is periodically bound by less mobile structures. If the substance diffuses rapidly but is significantly retarded by periodic binding to a static structure (such as the chromatin network), it can still exhibit an exponential decay of the fluorescence recovery, but the rate constant (k off ) will be largely determined by the mean residence time ( on ) on the networks. Approximately, k off ϭ 1/ on .
The startling first results of FRAP experiments with histone H1-GFP fusion appeared in two Nature papers published back-to-back (12,13). Both demonstrated that rather than being statically fixed on chromatin, most of the H1 was rapidly exchanging between sites, with mean residence time of only a few minutes (reviewed in Ref. 14.) In a comprehensive and careful analysis of a number of nuclear proteins genetically fused to GFP, Phair et al. (15) not only confirmed the earlier results on H1 but demonstrated rapid mobility for a wide class of nuclear proteins (see representative data in Table 1). In particular, this analysis showed that all of the proteins spent most of the time bound to static nuclear structures; nevertheless, all except the core histones were capable of quite rapid exchange; most exhibited bi-exponential exchange, suggesting two classes of binding sites.
The latter observation has not been fully explained but may result from the existence of both specific (slow exchanging) and nonspecific (fast exchanging) sites in the nucleus (15). Support for this interpretation comes from the observation that in two cases (H1°and HMGN1) mutations that are known to decrease affinity for specific nucleosomal binding essentially abolish the slow exchanging fraction (see Table 1).
Although the FRAP technique yields reproducible values for the mean residence time, it does not, at the moment, provide much information concerning the average time ( off ) the molecules spend in an unbound state, undergoing diffusion. An order of magnitude can be estimated by using the fact that the average fraction of bound molecules (f b ) is given by the equation, which can be rearranged to yield the following equation.
Given that most of the values of f b in Table 1 (last column) are close to unity, we can make only rough estimates of off . However, in all cases off Ͻ Ͻ on and generally is in the millisecond range. Thus, a typical molecule such as a transcription factor may be visualized as spending periods on the order of seconds on bound sites, alternated by millisecond periods of wandering about the nucleus. It has been estimated that one moderately sized protein molecule can, by free diffusion, traverse the entire nucleus in a matter of seconds (see Ref. 14). Even if the molecule spends only a few percent of the time in free diffusion, it will still be expected to pass such distances in a few minutes. Because there are many copies of most transcription factors, the chromatin of the nucleus can be thoroughly examined by each kind of protein in a brief period.
What does the mobility of the nuclear proteins signify with respect to nuclear processes? Before approaching this question, we must consider another newly recognized aspect of chromatin dynamism.

Fluctuations in Nucleosome Accessibility
Just as the local protein stoichiometry of chromatin was long thought to be mainly fixed, so was the structure of the nucleosome. Although it was realized that extremes of environmental conditions (temperatures, pH, ionic strength) could lead to nucleosome unfolding in vitro, it was long believed that such unfolding would occur under physiological conditions only in response to ATP-dependent "nucleosome remodeling" complexes.
Actually, the importance of inherent nucleosome instability has been suspected for many years: "in the physiological range of pH, ionic strength, and temperature the core particle is marginally stable. Stability appears to be increased by inclusion of H1 in the chromatosome and in the full chromatin structure. It seems reasonable that this incipient instability, which can be unlocked by removal of lysine-rich histones and by modification of the particles, is an essential feature of nucleosome function. An irreversibly condensed chromatin structure would be physiologically useless; what is seemingly required and provided is a structure that can exist in several levels of stability." (16).
However, the dynamic implications of this statement went unrealized until the seminal 1995 paper by Polach and Widom (17), which opened a whole new view of DNA accessibility in chromatin. In elegant, simple experiments, this group demonstrated that DNA sites even well within the putative nucleosomal coil could be, to a diminished extent, accessible to "reporter molecules" such as restriction endonucleases targeted to sites within the nucleosomal DNA. An important feature of the studies was the observation that accessibility monotonically decreased in progressing along the DNA from the nucleosome periphery toward the central dyad (see also Ref. 18 and Fig. 1). This result is precisely what might be expected from a spontaneous "unpeeling" of the DNA from the nucleosome core. In a further analysis, Widom and his colleagues (19) have shown how this model could account for "cooperative" binding of more than one ligand to the histone-DNA complex. Studies using prokaryotic RNA polymerase as a reporter molecule (20) set a lower limit of 0.13 s Ϫ1 for the unwrapping rate (corresponding to (closed) Х 7 s).
More recently, this same group has used fluorescence resonance energy transfer (FRET) to examine the postulated transition (21,22). In these experiments, one member of a fluorescent donor-acceptor pair was attached to the end of nucleosomal DNA, the other to a histone side chain that would be in proximity to this DNA end in the intact nucleosome, allowing efficient FRET (Fig. 2a). On the other hand, even modest unwrapping of the DNA should separate the donor and acceptor enough to significantly diminish energy transfer (see Fig. 2a).
In a first set of experiments, such nucleosomes were mixed in a stopped-flow cell with the protein Lex A, for which a binding site had been inserted in the nucleosomal DNA near the end to which the fluorescent donor was attached. A second set of experiments used fluorescence correlation spectroscopy (23) to compare fluorescence fluctuations from donor-only nucleosomes (arising from diffusion into and out of the field of view) to the fluctuations observed with nucleosomes containing both donor and acceptor. In the latter case, the fluctuations could arise from both diffusion and conformational changes that modified energy transfer. Together, these two kinds of FRET studies indicated average dwell times of about 0.25 s in the wrapped state and 0.01-0.05 s in the unwrapped state (22). This wrapped dwell time is at least an order of magnitude shorter than that reported by the same group using RNA polymerase as a probe (20). It may be that the polymerase requires a much more drastic unwrapping than does Lex A, which was arranged to bind close to the DNA exit.
Similar FRET studies by the group of K. Luger, using carefully defined and characterized nucleosomes, have further supported the idea that partial DNA unwrapping is a common phenomenon. For example, they have shown that binding of the transcription factor Amt1 causes partial unwrapping (24) and have characterized the effects of salt on nucleosome stability (25).
Ingenious as these experiments are, they remain as indirect measures of the unwrapping process and have been designed by the placement of donor and acceptor to be sensitive to even minor unwrapping. A quite different study by Tomschik et al. (26) has utilized single-molecule FRET experiments with  26. In a the donor dye is at the 5Ј end of the 147-bp sequence used for nucleosome reconstitution, and the acceptor fluorophore is attached to the unique cysteine introduced at position 35 of histone H3. The reconstituted nucleosome has a second acceptor dye on the second histone H3 molecule of the octamer, but it is situated at a distance of ϳ8 nm from the donor dye on the DNA and is not expected to contribute to the FRET signal. In b, both the donor and acceptor dyes are on the DNA 75 bp apart, deep inside the particle, so that changes in the efficiency of FRET would be observable only in the case of significant, long range unwrapping of the DNA from around the histone octamer. The right-hand panels in a and b depict the expected structures of the nucleosome when the FRET signal is lost, i.e. the DNA undergoes spontaneous detachment form the histone core to expose sites that would be hidden in the crystallographic conformation of the particle. Note that the positions of the donor and acceptor fluorophores in Refs. 21 and 26 would allow registering conformational fluctuation of nucleosomal DNA of different magnitudes.  nucleosomes having donor and acceptor groups placed deep into the nucleosomal DNA so that they are in close proximity near the dyad axis in the canonical wrapped particle (see Fig. 2b). For such molecules to show loss of energy transfer, a major unwrapping of the DNA from around the histone core would have to occur (Fig. 2b). Yet following individual nucleosomes in real time reveals that many molecules exhibit periods in which the FRET efficiency drops briefly to very low values (Fig. 3), corresponding to the unwrapping of at least 80 bp from the histone core. Analysis of a large number of such data recordings reveals dwell times in the closed state of several seconds, randomly interspersed by "open" periods averaging a few tenths of a second. It should be emphasized that these experiments sample a quite different set of structural fluctuations than those observed in the recent FRET experiments by the Widom group (21,22). The latter, because of the locations of donor and acceptor, can respond to a much lower degree of unwrapping then the construct used by Tomschik et al. (26). Simple geometrical calculations show that the Widom construct should respond by a major drop in FRET by an unpeeling of only ϳ25 bp, where ϳ80 bp is needed for a comparable change with the Tomschik construct (Fig. 2). Nevertheless, it is of interest that the most extreme and infrequent site exposures inferred from the early enzyme accessibility studies (17, 18) (see Fig. 1) correspond to those that have been seen using the Tomschik et al. (26) construct. Such extreme "all or none" unwrapping was in fact anticipated many years ago in a theoretical analysis (27). The important point is that different research groups, using a variety of techniques, have provided evidence that the nucleosome is not a static structure but one in which the DNA periodically exhibits significant unwrapping from the histone core. That this involves DNA unwrapping, rather than some sort of histone octamer unfolding, is evidenced by experiments using crosslinked octamers, which yield the same results (26). What are the likely consequences for chromatin function? How does this relate to nuclear protein mobility? To approach these questions, we must compare the time scales characteristic of protein mobility and nucleosome fluctuation.

Gated Sites and Stochastic Sampling
The behavior of a nucleosome, which can periodically expose part of its DNA to an environment containing potential binding proteins, is formally identical to that of a "gated" enzyme, in which the active site is only occasionally exposed to substrate. There exists a large body of theory directed toward such entities (see, for example, Refs. 28 -30).
A major feature of such interactions is that the control of binding to such a site depends critically on the ratio of the dwell time in the closed state to a characteristic diffusional relaxation time, defined as ϭ R 2 /D, where R is the radius of the gated enzyme, and D is the sum of the diffusion coefficients of enzyme and ligand (28). We consider two limiting cases, one in which dwell times are very long compared with , and the opposite. If the closed dwell times are Ͼ Ͼ, the effective binding rate is proportional to the fraction of time the gates are open, as we might expect, but if the closed dwell times are very short, the system can behave as if the gate were always open; the binding rate is the same as expected for an ungated reaction. This surprising result can be explained in the following way: if gates are opening and closing rapidly, as compared with the time required for significant diffusional motion, every ligand that is near a gate will remain there long enough to have one or more opportunities to enter. The consequence of all this is the following: only if the gate opening/closing rate is slow compared with ligand diffusion will modification of the gating function affect the accessibility of ligands to the site.
What is the situation with respect to nucleosome opening/closing and ligand binding? Accurate estimation of the diffusion relaxation time for a moderately sized protein approaching a gated nucleosome is difficult because of approximations in the theory. However, even the approximate values predicted are so small (in the microsecond range) as compared with the observed dwell times in the closed state (in the seconds range) that there is no question: we are in the "slow gating" limit. Thus, anything that modifies either the fraction of time the nucleosome stays open, or the opening/closing rates, can potentially modify the rates of binding of transcription factors, etc. It will be fascinating to learn whether or not covalent modification or variation of nucleosomal histone composition can have such effects. Preliminary measurements by the enzyme accessibility assay indicate that at least with histone acetylation small, but measurable, changes in accessibility are seen (31).
One factor that should have a major effect on nucleosome gating is the presence of linker histones, which appears to strongly inhibit the opening process; a very specific example is given in Ref. 32. However, the inhibition is not complete, probably because of the transient nature of H1 binding, as described in the preceding section. This brings us to consider the intimate interrelationship between protein mobility in the nucleus, nucleosome gating, and chromatin function.

Dynamic Nuclear Processes, an Integrated View
The parallel discoveries of nuclear protein mobility and dynamic nucleosome unwrapping converge to provide a new conceptual framework for processes like transcriptional activation. The general scheme is indicated in Fig. 4. In this view, many of the steps leading up to the formation of a multiprotein functional complex on chromatin are seen to be in principle reversible. In addition, any nucleosome carrying the appropriate factor binding sites is, in principle, accessible to binding. A necessary first step is the dissociation of H1,  (26). The changes in the fluorescence intensities of the donor (Cy3, green) and acceptor (Cy5, red) fluorophores bound to a single nucleosome (construct in Fig. 2b) have been recorded as a function of time. The example trace shows evidence of nucleosomal dynamics: the particle is either in a closed conformation characterized by high efficiency of FRET (high acceptor signal and low donor signal) or in an open conformation characterized by low FRET (low acceptor signal and high donor signal). The sudden drop in fluorescence intensity of the acceptor Cy5 is due to photobleaching, i.e. loss of the ability of the dye to be excited. The conformational transitions inferred on the basis of the measured efficiency of FRET are depicted in Fig. 2b. The open state is accessible to more proteins, including nucleosome remodeling factors, leading to the assembly of functional macrostructures such as the transcription initiation complex. The schematic also depicts a scenario of formation of a more stable "closed" state, which is achieved by binding of the so-called "closing" factors at the entry-exit nucleosomal site. These factors have higher binding affinity than the majority of linker histone variants and may keep the particle closed for lengthy periods of time. Examples of such factors may include the nucleated erythrocyte-specific linker histone variant H5 (which binds to chromatin irreversibly in terminally differentiated erythrocytes that are inactive in replication and transcription) or the methyl-DNA-binding protein MeCP2 (for a recent review, see Ref. 49). The binding of closing factors can occur in the open state (depicted), or alternatively such factors may actively displace linker histones, as seems to be the case for MeCP2. preceded or accompanied by opening of the higher order structure. There is evidence that the opening of the higher order structure is itself a spontaneous process (33), with the opened state being stabilized by specific proteins (34). However, opening of the higher order structure may not in fact be essential, since recent reports strongly argue that condensed chromatin domains, even mitotic chromosomes, are easily accessible to large macromolecules (35,36). The release of H1 can be spontaneous and may also be under cellular control, for it has been noted that reagents modifying H1 phosphorylation also modify the dynamics of H1 exchange (12,14).
The H1-depleted nucleosome may either rebind another H1 or undergo an unwrapping event. If the latter happens, a number of possible scenarios follow (see Fig. 4). For example, the partially unwrapped nucleosome may accept one or more molecules of an "opening factor"; if it does, complete re-wrapping should be inhibited. Possible examples of such factors are proteins of the HMG class; abundant evidence indicates that these proteins bind to the DNA near the nucleosome entry/exit points (see Refs. 37 and 38). By binding in this manner, HMG proteins could prop open the "crossover" of DNA near the entry/exit region, a structure that seems to be involved in H1 binding (39); as a consequence, H1 rebinding should be blocked when HMG is present. Thus, there is expected to be a dynamic equilibrium between H1 and HMG proteins, as noted by others (see, for examples, Refs. [37][38][39][40]. In any event, up to this point the steps depicted in Fig. 4 should be reversible. Dissociation of HMG could (but not necessarily would) be followed by H1 rebinding. The importance of the HMG is that, if present in high concentrations, it may serve to "block open" nucleosomal DNA regions, in a nonspecific manner, for further interaction. Such interaction might involve either binding of additional transcription factors or histone modifying enzymes or a chromatin-remodeling complex. Support for such a role for HMG proteins is provided by the recent observation that HMGB1 increases the residence time of the glucocorticoid receptor on chromatin (41). For discussion of likely sequences of such further events in the assembly of the initiation complex, see Refs. 6 and 42. At some point, as more and more protein components are added to the complex, we may presume that the binding becomes essentially irreversible. If the region concerned includes a promoter site, the ultimate step will be the completion of the preinitiation complex, culminating in transcription (see Ref. 43 for a recent overview of the process). The idea that dynamic nucleosome opening may provide access for transcriptional activators is hardly new; it is elegantly expressed, with examples, in a recent review by Morse (44). However, combination of this with recent information on nuclear protein mobility does seem to open new vistas.
The model has the attractive aspect that no preselection or "marking" of a particular promoter region is necessary. All promoters are potentially accessible, and one may expect that each is sampled more than once during the life of a cell. How broad and general this sampling can be is indicated by the recent observations that HP1, a protein associated with heterochromatic regions, is itself mobile and dynamic in its binding (45,46). Which sites are ultimately chosen for initiation depends on the presence (or more precisely, concentrations) of a number of transcription factors, coactivators, etc. At each of the later steps in the stochastic process depicted in Fig. 4, the probability of proceeding to the next step depends critically on the concentration of the next participant. This will produce a multiplicative effect and an enormous advantage for genes favored by the nuclear content of the appropriate factors because all factors must be available in sufficient concentration to produce a given final result. Furthermore, organismal development can proceed without fundamental changes in chromatin structure by simply changing the concentrations of critical proteins.

Where from Here?
The implications of the new paradigm have hardly been sounded. There is much that we still do not know concerning either of the two basic observations, protein mobility and nucleosome accessibility. Why do so many nuclear proteins exhibit two (or more) components in relaxation? Does this represent (as suggested in Ref. 15) two classes of binding sites, and if so, what are they? Why do so many proteins, of very different functions, exhibit such similar behavior? To what extent does simple physical obstruction by chromatin, by nuclear matrix, or both contribute to the retardation of diffusion? Is mobility different in different nuclear compartments? The list of questions is almost endless.
A similar situation exists with regard to the dynamics of nucleosomes. Are there different classes of nucleosomes with different unwrapping behavior? The single-molecule measurements (26) revealed only a portion of the population undergoing FRET fluctuations. Do these represent different classes (involving different histone variant compositions, for example), or is this an artifact of the experiment? Are there histone modifications that strongly modulate dynamic behavior? What will be observed in the presence of HMG proteins or transcription factors?
Finally, we must be wary. All of the published studies of nucleosome dynamics have been in vitro. There is as yet no evidence that such transitions occur in chromatin in vivo. It has only very recently been shown that internal sites within oligonucleosomal arrays can be assessed by nucleases (47).
Once again, new techniques and a new perspective have made a research area that was becoming moribund appear fresh and exciting. The next several years should be very interesting.